Elon Musk is pushing ahead with plans to deploy artificial intelligence data centers in orbit as early as 2027, a radical bet that he claims will solve the power bottlenecks constraining AI on Earth. The SpaceX chief executive’s proposal has opened a new front in an already fractious debate over the future of AI infrastructure, drawing sharp skepticism from rivals including SoftBank Group Corp. (9984.T) founder Masayoshi Son and OpenAI CEO Sam Altman.

Musk’s vision, detailed in recent investor presentations and public posts on X, revolves around a satellite constellation called Starmind. Each spacecraft would carry a 150-kilowatt compute payload powered by a 150-kilowatt solar array spanning 70 meters, linked by laser connections and beaming processing capacity back to the ground through Starlink. To manage the intense heat, the satellites would use deployable liquid radiators that dump thermal energy directly into the vacuum of space, bypassing the water and air-cooling constraints that hobble terrestrial facilities.

“Space is the only way to scale at scale,” Musk wrote on X, framing Earth-based power grids and cooling systems as insurmountable limits for the next generation of AI models. In a separate post the same day, he told a critic that “SpaceX will be worth more than the rest of Earth if we accomplish our goals,” a statement that implies a valuation exceeding the roughly $110 trillion in annual global economic output.

The timing was no coincidence. SpaceX published the first public renderings of its AI1 satellite that day, giving investors a concrete look at the hardware behind the Starmind concept. Musk has been selling the near-term version of the plan to SpaceX backers as a way to capture demand that he insists ground-based data centers cannot meet.

Not everyone is buying it. OpenAI’s Altman has dismissed the pitch as premature, according to a report by 《Nikkei》, saying Musk is “selling investors on a short-term space data center concept” that ignores the brutal economics of operating compute in orbit. Independent estimates cited by the Japanese newspaper peg the operating cost of a space-based data center at roughly five times that of an equivalent ground facility, a gap driven by launch expenses, radiation hardening, and the complexity of servicing hardware hundreds of miles above the planet.

SoftBank’s Son has taken a different tack. Rather than moving servers to space, Son has championed a sprawling terrestrial buildout, arguing that the fastest path to AI scale is pouring capital into land, power and cooling on the ground. The divergence in strategy matters because the three men—Musk, Altman and Son—are effectively competing to define the infrastructure layer on which the entire AI industry will run.

The $40 Billion Customer

Beneath the orbital ambitions, Musk’s AI maneuvering has a more immediate financial dimension. In May, Anthropic agreed to lease the full output of xAI’s Colossus 1 data center near Memphis, roughly 300 megawatts of capacity, at a cost of about $1.25 billion a month through 2029. That deal will channel close to $40 billion to Musk’s side of the table.

The commercial relationship helps explain a striking public pivot. Musk, who in February called Anthropic’s models “misanthropic and evil,” now describes the company as the “clear leader in AI.” Responding on X to a user who noted that SpaceX controls the computing power Anthropic depends on and could theoretically cut it off, Musk pledged he never would, adding that no rival had shipped a model as good as Anthropic’s Mythos and Fable systems and that he expects a Mythos 2 release soon.

“A rival who pays you $40 billion is a rival worth flattering,” one industry observer noted. The praise costs Musk nothing and buys goodwill with a customer whose rent payments are material to xAI’s revenue. It also serves a second purpose: elevating Anthropic at the expense of OpenAI, the company Musk co-founded in 2015, left in 2018 after failing to win control, and has fought ever since.

Musk sued Altman and OpenAI in 2024, seeking more than $150 billion in damages and demanding the unwinding of the for-profit structure. A jury threw the case out in May on statute-of-limitations grounds, a ruling Musk called a “calendar technicality” and vowed to appeal. By July, the feud had turned personal again. Musk branded Altman “Scam Altman” after Apple sued OpenAI, and Altman replied that the surest sign his new model led the field was that Musk was obsessed with him again.

A Race to the Public Markets

The personal and financial threads are converging around a pair of blockbuster IPOs. Both Anthropic and OpenAI filed confidentially for stock market listings in June, within days of each other. Anthropic, valued at roughly $965 billion in private markets, is pushing for a Nasdaq debut as early as October and expects its first profitable quarter, with around $559 million in operating income on $10.9 billion of revenue. OpenAI, valued a little lower and still loss-making, is leaning toward a 2027 listing, wary after SpaceX’s own IPO spiked and then surrendered much of its gain. Altman is holding out for a $1 trillion valuation.

Whoever lists first sets the benchmark against which the other will be priced. Anthropic going out ahead, blessed by Musk as the field’s leader, helps fix the multiple that bankers will later apply to OpenAI. Musk talking up the rival he profits from, while running down the rival he is suing, shapes the terms on which his enemy will one day face public investors.

CompanyEstimated Private ValuationIPO TimelineRecent Financial SnapshotAnthropic~$965 billionAs early as October 2026First profitable quarter expected; ~$559M operating income on $10.9B revenueOpenAIBelow AnthropicTargeting 2027Still loss-making; CEO seeking $1 trillion valuationSpaceX~$400 billionListed; post-IPO volatilityStarmind constellation and AI1 satellite revealed July 2026

Note: Financial figures for Anthropic are based on pre-IPO expectations and have not been publicly verified through regulatory filings.

None of this proves Musk’s praise for Anthropic is hollow. He may genuinely believe the company makes the best models today. But the compliment lands in the narrow window before both labs ask the market to price them, serving his balance sheet and his grudge at once.

The Orbital Bet

Musk’s space data center concept, while years from commercial operation, is already shaping how investors think about SpaceX’s long-term value. The company is currently valued at around $400 billion. Musk’s assertion that it could one day exceed the rest of Earth’s combined output is a characteristically maximalist framing, but the underlying logic is that a company controlling the infrastructure of a space-based economy—one that can harness energy and manufacturing potential beyond the atmosphere—could eventually operate at a scale that makes terrestrial GDP look small.

The Starmind constellation is the first tangible step toward that vision. Each AI1 satellite’s 150-kilowatt compute payload is modest compared with a ground-based hyperscale data center, but the architecture is designed to scale horizontally. Laser-linked satellites can function as a distributed supercomputer, and the absence of land, grid and water constraints means the theoretical ceiling is far higher than anything achievable on Earth.

Skeptics point to the five-to-one cost disadvantage cited in the 《Nikkei》 report, as well as the latency and bandwidth challenges of beaming compute from low Earth orbit. Proponents counter that launch costs have fallen dramatically and that the power bottleneck on the ground is worsening faster than most forecasts anticipated. The debate is unlikely to be resolved until hardware is in orbit and serving paying customers.

What is already clear is that the battle over AI infrastructure has moved beyond data center square footage and gigawatt power deals. It now extends to orbit, to the courtroom, and to the IPO calendar, with Musk positioned at the center of every dimension.